If gpio_chip->base is negative during registration, gpiolib requests
dynamic base allocation. This is useful for devices being registered
at run-time (in contrast to platform devices).
Signed-off-by: Anton Vorontsov <[email protected]>
---
drivers/gpio/gpiolib.c | 42 ++++++++++++++++++++++++++++++++++++------
1 files changed, 36 insertions(+), 6 deletions(-)
diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
index d8db2f8..1d25104 100644
--- a/drivers/gpio/gpiolib.c
+++ b/drivers/gpio/gpiolib.c
@@ -77,6 +77,32 @@ static inline struct gpio_chip *gpio_to_chip(unsigned gpio)
return gpio_desc[gpio].chip;
}
+static int gpiochip_find_base(int ngpio)
+{
+ int i;
+ int spare = 0;
+ int next_base = 0;
+ int base = -ENOSPC;
+
+ for (i = 0; i < ARCH_NR_GPIOS; i++) {
+ struct gpio_chip *c = gpio_desc[i].chip;
+
+ if (!c) {
+ spare++;
+ if (spare == ngpio) {
+ base = next_base;
+ break;
+ }
+ } else {
+ spare = 0;
+ i += c->ngpio;
+ next_base = i;
+ }
+ }
+
+ return base;
+}
+
/**
* gpiochip_add() - register a gpio_chip
* @chip: the chip to register, with chip->base initialized
@@ -92,17 +118,21 @@ int gpiochip_add(struct gpio_chip *chip)
int status = 0;
unsigned id;
- /* NOTE chip->base negative is reserved to mean a request for
- * dynamic allocation. We don't currently support that.
- */
-
- if (chip->base < 0 || (chip->base + chip->ngpio) >= ARCH_NR_GPIOS) {
+ if (chip->base >= 0 && chip->base + chip->ngpio >= ARCH_NR_GPIOS) {
status = -EINVAL;
goto fail;
}
spin_lock_irqsave(&gpio_lock, flags);
+ if (chip->base < 0) {
+ chip->base = gpiochip_find_base(chip->ngpio);
+ if (chip->base < 0) {
+ status = chip->base;
+ goto fail_unlock;
+ }
+ }
+
/* these GPIO numbers must not be managed by another gpio_chip */
for (id = chip->base; id < chip->base + chip->ngpio; id++) {
if (gpio_desc[id].chip != NULL) {
@@ -116,7 +146,7 @@ int gpiochip_add(struct gpio_chip *chip)
gpio_desc[id].flags = 0;
}
}
-
+fail_unlock:
spin_unlock_irqrestore(&gpio_lock, flags);
fail:
/* failures here can mean systems won't boot... */
--
1.5.2.2
On Tue, 4 Mar 2008 20:03:07 +0300
Anton Vorontsov <[email protected]> wrote:
> If gpio_chip->base is negative during registration, gpiolib requests
> dynamic base allocation. This is useful for devices being registered
> at run-time (in contrast to platform devices).
>
> Signed-off-by: Anton Vorontsov <[email protected]>
> ---
> drivers/gpio/gpiolib.c | 42 ++++++++++++++++++++++++++++++++++++------
> 1 files changed, 36 insertions(+), 6 deletions(-)
>
This conflicts in non-trivial ways with gpio-define-gpio_is_valid.patch.
Could you please redo and retest against 2.6.25-rc3-mm1?
Thanks.
On Wednesday 05 March 2008, Andrew Morton wrote:
> On Tue, 4 Mar 2008 20:03:07 +0300
> Anton Vorontsov <[email protected]> wrote:
>
> > If gpio_chip->base is negative during registration, gpiolib requests
> > dynamic base allocation. This is useful for devices being registered
> > at run-time (in contrast to platform devices).
> >
> > Signed-off-by: Anton Vorontsov <[email protected]>
> > ---
> > drivers/gpio/gpiolib.c | 42 ++++++++++++++++++++++++++++++++++++------
> > 1 files changed, 36 insertions(+), 6 deletions(-)
> >
>
> This conflicts in non-trivial ways with gpio-define-gpio_is_valid.patch.
> Could you please redo and retest against 2.6.25-rc3-mm1?
And when you do that, could you make it assign GPIO numbers from
the biggest number on down, instead of from the smallest on up?
Platforms normally assign those numbers from the bottom up ... so
dynamic assignment should try to avoid reusing any numbers that may
have been explicitly assigned, but not yet registered. (Of course,
if you can come up with a cleaner solution than that, it'd be great.
Maybe platforms should also be able to mark ranges as "in use" as
they start up, or something.)
And please add a pr_debug level message reporting dynamically
assigned ranges. That way, when problems crop up it'll be that
much easier to notice what went wrong.
- Dave
On Wed, Mar 05, 2008 at 03:40:57PM -0800, David Brownell wrote:
> On Wednesday 05 March 2008, Andrew Morton wrote:
> > On Tue, 4 Mar 2008 20:03:07 +0300
> > Anton Vorontsov <[email protected]> wrote:
> >
> > > If gpio_chip->base is negative during registration, gpiolib requests
> > > dynamic base allocation. This is useful for devices being registered
> > > at run-time (in contrast to platform devices).
> > >
> > > Signed-off-by: Anton Vorontsov <[email protected]>
> > > ---
> > > drivers/gpio/gpiolib.c | 42 ++++++++++++++++++++++++++++++++++++------
> > > 1 files changed, 36 insertions(+), 6 deletions(-)
> > >
> >
> > This conflicts in non-trivial ways with gpio-define-gpio_is_valid.patch.
> > Could you please redo and retest against 2.6.25-rc3-mm1?
No problem, will rebase.
> And when you do that, could you make it assign GPIO numbers from
> the biggest number on down, instead of from the smallest on up?
>
> Platforms normally assign those numbers from the bottom up ... so
> dynamic assignment should try to avoid reusing any numbers that may
> have been explicitly assigned, but not yet registered. (Of course,
> if you can come up with a cleaner solution than that, it'd be great.
> Maybe platforms should also be able to mark ranges as "in use" as
> they start up, or something.)
Well, it is also trivial to implement gpiochip_reserve(start, end)
function that will mark reserved gpio->chip as ERR_PTR(-EACCESS),
and these ranges will be unavailable for the dynamic allocation.
Then, platforms could do gpiochip_reserve(0, GPIO_ARCH_END).
I think This is good solution from all and every POV, but this
needs a bit of platform code assistance.
What would you choice?
> And please add a pr_debug level message reporting dynamically
> assigned ranges. That way, when problems crop up it'll be that
> much easier to notice what went wrong.
Will do.
Thanks,
--
Anton Vorontsov
email: [email protected]
irc://irc.freenode.net/bd2